A network may be defined by the pattern of connections between elements of the network, commonly referred to as the network topology. Data centers often use a multi-stage Clos or folded-Clos network topology, in which a network element may be linked to multiple higher-level network elements: servers are connected to top-of-rack (ToR) switches, each ToR switch is connected to multiple switches of the next layer, which in turn are connected to multiple next-tier switches, and so on. Such a topology provides both high bandwidth and a high tolerance for path failures, and is therefore excellently suited to periods of high network demand.
But such a topology may be over-provisioned. Actual network demand is often much lower than peak demand, but the servers and switches in the network will draw largely as much power when they have no demand to respond to as they will when working at full capacity. A data center must be able to meet the highest level of demand for its services, but it should minimize the overall power consumption while providing resiliency.